[Paper Review] From Ecological Connectivity to Outbreak Risk: A Heterogeneous Graph Network for Epidemiological Reasoning under Sparse Spatiotemporal Data

Haley Stone, Jing Du+9 more|arXiv (Cornell University)|Jan 8, 2026

Zoonotic diseases and public health0 citations

TL;DR

zooNet is a graph-based framework that fuses ecological transmission simulation, metadata-driven genetic distance imputation, and spatiotemporal graph learning to infer outbreak dynamics of A/H5 avian influenza in the US under sparse surveillance.

ABSTRACT

Estimating population-level prevalence and transmission dynamics of wildlife pathogens can be challenging, partly because surveillance data is sparse, detection-driven, and unevenly sequenced. Using highly pathogenic avian influenza A/H5 clade 2.3.4.4b as a case study, we develop zooNet, a graph-based epidemiological framework that integrates mechanistic transmission simulation, metadata-driven genetic distance imputation, and spatiotemporal graph learning to reconstruct outbreak dynamics from incomplete observations. Applied to wild bird surveillance data from the United States during 2022, zooNet recovered coherent spatiotemporal structure despite intermittent detections, revealing sustained regional circulation across multiple migratory flyways. The framework consistently identified counties with ongoing transmission weeks to months before confirmed detections, including persistent activity in northeastern regions prior to documented re-emergence. These signals were detectable even in areas with sparse sequencing and irregular reporting. These results show that explicitly representing ecological processes and inferred genomic connectivity within a unified graph structure allows persistence and spatial risk structure to be inferred from detection-driven wildlife surveillance data.

Motivation & Objective

• Motivate estimation of population-level prevalence and transmission in wildlife under sparse, biased surveillance data.
• Develop a modular framework that fuses ecological, genomic, and epidemiological information into a unified graph representation.
• Enable inference of outbreak connectivity and transmission dynamics despite incomplete observations.
• Assess spatial and temporal patterns of A/H5 spread across US migratory flyways.

Proposed method

• Construct a bi-layer heterogeneous graph with outbreak events and administrative regions.
• Augment detections with SEI-based ecological simulations to generate synthetic infections.
• Impute genetic distances for unsequenced or augmented cases via a metadata-driven quantile regression model using the K80 distance.
• Use cross-layer smoothing and graph fusion to integrate ecological, genetic, and spatial relations while preserving graph properties.
• Model temporal dynamics with an autoregressive graph encoder–decoder over successive time steps.
• Evaluate performance across US avian flyways with cross-validation and compare to baselines.

Figure 1: zooNet integrates ecological simulation, genetic distance imputation, and multilayer graph learning to forecast the spread of avian influenza A/H5 in wild birds. SEI-based simulation generates synthetic infections using environmental and host factors. Genetic distances are imputed using a

Experimental results

Research questions

• RQ1 Can zooNet recover coherent spatiotemporal structure of A/H5 outbreaks under sparse, detection-driven surveillance?
• RQ2 How do ecological connectivity, genetic connectivity, and spatial correlations combine to predict outbreak risk across flyways?
• RQ3 What is the predictive performance and signal-utility of incorporating SEI augmentations and genetic-distance imputation?
• RQ4 How does performance vary across migratory flyways and over the course of the surveillance season?

Key findings

• zooNet produced coherent spatiotemporal patterns of A/H5 risk across US migratory flyways despite intermittent detections.
• Predictive performance varied by flyway, with MSE lowest in the Atlantic (49.85 ± 10.48) and highest in the Pacific (155.81 ± 48.91).
• F1 scores of detections ranged from 0.0565 ± 0.0233 (Central) to 0.1383 ± 0.0725 (Pacific), indicating higher recall than precision.
• Pearson correlations between predicted and observed trends were highest in the Pacific (0.2173 ± 0.0644) and lowest in the Central flyway (0.0950 ± 0.0280).
• Ablation showed the full zooNet generally outperformed variants; SEI augmentation and genetic distance imputation contributed variably by flyway.
• The framework captures seasonal and spatial heterogeneity, with predicted outbreaks anteceding some confirmed detections in several counties.

Figure 2: Effect of module removal on prediction error and standard deviation across flyways.